基于图像处理的铁轨表面缺陷检测设计

发布时间：2018-03-03 20:31

本文选题：图像滤波　切入点：图像分割　出处：《西南科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着我国铁路高速发展,传统的铁路轨道检测手段已无法满足高效、准确的检测需要。本文以铁轨图像为研究对象,结合图像处理、模式识别等相关技术,提出一套基于图像处理的铁轨表面缺陷检测系统,该系统能够实现对铁轨表面缺陷区域进行智能识别。本文主要包括以下几方面内容。首先,通过对比几种传统的图像滤波算法,采用一种结合均值的快速中值滤波方法对图像进行初步的滤波处理;结合形态顶帽运算,提出一种基于梯度投影的方法对铁轨表面区域进行定位,通过定位结果验证该方法的有效性。其次,通过对比前后图像的差异,来初步判断图像是否存在缺陷;结合人类视觉注意机制,采用视觉显著性计算方法对缺陷区域进行分割,得到准确的缺陷区域,并与SR算法和Itti算法得到的结果做对比性实验;对缺陷区域进行特征提取,包括缺陷部分的形状特征、几何特征、灰度特征和纹理特征,分析提取到的数据,选取合适的特征训练SVM分类器并完成SVM核函数参数优化,实现缺陷与伪缺陷(轨道连接处)的分类。最后,设计缺陷检测软件来实现上述算法,并对软件的可靠性进行检测。经过实验验证,本文算法具有良好的稳定性,经过统计最终的检测识别率为91.67%,在实际应用中具有参考价值。
[Abstract]:With the rapid development of railway in China, the traditional railway track detection methods can not meet the needs of efficient and accurate detection. This paper takes rail image as the research object, combines image processing, pattern recognition and other related technologies. An image processing based rail surface defect detection system is proposed. The system can realize the intelligent recognition of the rail surface defect area. This paper mainly includes the following aspects. By comparing several traditional image filtering algorithms, a fast median filtering method combined with mean value is used to process the image. A method based on gradient projection is proposed to locate the rail surface area, and the results show that the method is effective. Secondly, the defect of the image is preliminarily judged by comparing the difference of the image before and after. Combined with the mechanism of human visual attention, the defect region is segmented by visual salience calculation method, and the accurate defect region is obtained, and compared with the results obtained by SR algorithm and Itti algorithm, the defect region is extracted by feature extraction, and compared with the results obtained by SR algorithm and Itti algorithm. It includes shape feature, geometry feature, grayscale feature and texture feature of defect part, analyzes the extracted data, selects suitable features to train SVM classifier and optimizes the parameters of SVM kernel function. Finally, the defect detection software is designed to realize the above algorithm, and the reliability of the software is tested. The experimental results show that the algorithm has good stability. The final detection and recognition rate is 91.67, which has reference value in practical application.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U216.3;TP391.41

【相似文献】